带模糊语义的神经网络规则提取  

Extracting Symbolic Rules from Neural Networks with Fuzzy Concepts

作　　者：张向华[1] 

机构地区：[1]重庆大学计算机学院

出　　处：《重庆大学学报（自然科学版）》2008年第3期328-331,共4页Journal of Chongqing University

基　　金：重庆市教委基金资助项目(KJ071504)

摘　　要：针对传统的规则提取方法在处理连续值输入属性时带有很大的盲目性,且其描述也不符合人类的认知习惯的弊端,在对比原有方法的基础上,引入模糊语义,提出了新的处理连续值函数的方法,从神经网络中提取出带模糊语义的符号规则,提高了规则的可理解性。因此,使用者可以很方便地验证它的正确性。通过把连续值神经网络转化成二值网络,利用二值网络布尔规则提取方法来提取带模糊语义的规则,更符合人们的思维习惯。The extracting symbolic rules from neural networks（NNs） becomes the bridge for expert system and NNs, and a main tool to solve the comprehensibility of NNs. The traditional rules to solve the problem show quite uncertain in the processing of continuous value, and furthermore this does not accord with human＇s habit. We present a new method with fuzzy concepts to solve the continuous value based on the traditional methods, which improves the comprehensibility of rules. So users can have a thorough understanding of the dealing process of system problem and validate its correctness. By transforming consecutive neural network into two-value network, users can utilize its regular extract method to extract rules with fuzzy concepts, which is more conform to thinking habits of human beings.

关 键 词：神经网络 规则提取 模糊语义 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]